search

ZhengyaoJiang / PGPortfolio

PGPortfolio: Policy Gradient Portfolio, the source code of "A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem"(https://arxiv.org/pdf/1706.10059.pdf).
GNU General Public License v3.0
1.74k stars 750 forks source link

Sharpe ratio as loss function #33

Closed lytkarinskiy closed 6 years ago

lytkarinskiy commented 6 years ago

Hello all!

I wonder if you've tried to use Sharpe ratio directly to optimize portfolio selection? With controlling both mean and variance parts it seems to allow control risk management of portfolio selection and minimize drawdawn?

Although I saw in literature that risk sensitive policy is less effective than risk neutral one I want to hear your opinion.

Thanks!

ZhengyaoJiang commented 6 years ago

Greetings, This question is related to our recent unpublished work. I'm sorry that I cannot discuss too much the topic. However, what I can tell is using Sharpe ratio directly here won't work.

Regards zhengyao

lytkarinskiy commented 6 years ago

Wow, besides the work already've been done it's a huge next step! Can you tell any possible date we can read the paper? =)

Andrey

ZhengyaoJiang commented 6 years ago

I think next year. Not sure about the exact date yet.

Zhengyao

akaniklaus commented 6 years ago

@lytkarinskiy Found some example implementations below, why don't you try to integrate that for us? https://github.com/TenzinDJ/AI_for_Finance https://github.com/SergeOlivierP/agentF https://github.com/OlegMitsik/AA228_Project